Browsing by Author "Wu, JM"

Now showing 1 - 3 of 3
  • No Thumbnail Available
    Item
    Identifying chemical factors affecting reaction kinetics in Li-air battery via ab initio calculations and machine learning
    (Elsevier, 2021-03-01) Wang, AP; Zou, ZY; Wang, D; Liu, Y; Li, YJ; Wu, JM; Avdeev, M; Shi, S
    Redox mediators are promised to thermodynamically resolve the cathode irreversibility of Li-air battery. However, the sluggish chemical reaction between mediators and discharge products severely restrains fast charging. Here, we combine ab initio calculations and machine learning method to investigate the reaction kinetics between LiOH and I2, and demonstrate the critical role of the disorder degree of LiOH and the solvent effect. The Li+ desorption is identified as the rate determining step (rds) of the reaction. While LiOH turns from the crystalline to disordered/amorphous structure, the rds energy barrier will be reduced by ∼500 meV. The functional group of the solvent is detected as the key to regulating the solvation effect and phosphate-based solvent is predicted to accelerate the decomposition kinetics most with the strongest solvation capability. These findings indicate that the faster reaction kinetics between mediators and the discharge products can be achieved by rational discharge product structure regulation and appropriate solvent selection. © 2020 Elsevier B.V.
  • Thumbnail Image
    Item
    Multi‐layer feature selection incorporating weighted score‐based expert knowledge toward modeling materials with targeted properties
    (Wiley, 2020-01-15) Liu, Y; Wu, JM; Avdeev, M; Shi, SQ
    Selecting proper descriptors or features is one of the central problems in exploring structure–activity relationships of materials using machine learning models. The current feature selection algorithms usually require tedious hyperparameter tuning and do not actively consider the prior knowledge of domain experts about the features. Here, this work proposes a data‐driven multi‐layer feature selection method incorporating domain expert knowledge named DML‐FSdek, which is automated, with users entering training data without manual tuning of the hyperparameters. The domain expert knowledge is quantified by means of weighted scoring and integrated into the selection process to eliminate the risk of crucial features being removed. The test studies on ten material properties datasets demonstrate the potential of the approach to automatically search for a reduced feature set with lower root mean square errors than those for the initial feature set. Essentially, the most relevant material features, the number of which is much smaller than that in the original feature set, are automatically selected to establish a closer and more accurate structure–activity relationship for the materials of interest. As a result, the method represents the targeted properties of materials with a smaller and more interpretable set of features while ensuring equal or better prediction accuracy. © 2020 The Authors. Published by WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim
  • No Thumbnail Available
    Item
    Predicting creep rupture life of Ni-based single crystal superalloys using divide-and-conquer approach based machine learning
    (Elsevier, 2020-05-17) Liu, Y; Wu, JM; Wang, ZC; Lu, XG; Avdeev, M; Shi, S; Wang, CY; Yu, T
    Creep rupture life is a key material parameter for service life and mechanical properties of Ni-based single crystal superalloy materials. Therefore, it is of much practical significance to accurately and efficiently predict creep life. Here, we develop a divide-and-conquer self-adaptive (DCSA) learning method incorporating multiple material descriptors for rational and accelerated prediction of the creep rupture life. We characterize a high-quality creep dataset of 266 alloy samples with such features as alloy composition, test temperature, test stress, and heat treatment process. In addition, five microstructural parameters related to creep process, including stacking fault energy, lattice parameter, mole fraction of the γ' phase, diffusion coefficient and shear modulus, are calculated and introduced by the CALPHAD (CALculation of PHAse Diagrams) method and basic materials structure-property relationships, that enables us to reveal the effect of microstructure on creep properties. The machine learning explorations conducted on the creep dataset demonstrate the potential of the approach to achieve higher prediction accuracy with RMSE, MAPE and R2 of 0.3839, 0.0003 and 0.9176 than five alternative state-of-the-art machine learning models. On the newly collected 8 alloy samples, the error between the predicted creep life value and the experimental measured value is within the acceptable range (6.4486 h–40.7159 h), further confirming the validity of our DCSA model. Essentially, our method can establish accurate structure-property relationship mapping for the creep rupture life in a faster and cheaper manner than experiments and is expected to serve for inverse design of alloys. © 2020 Acta Materialia Inc. Published by Elsevier Ltd.